Novel mobile robot FastSLAM based on unscented transform

This paper proposes a novel FastSLAM based on UT (Unscented Transform). A novel proposal distribution which integrates current observation is proposed, wherein the particles sampled from the proposal distribution based on Unscented Transform is driven to move to the high probability region of posterior distribution. The landmarks in map are updated with UKF (Unscented Kalman Filter) to avoid the problem of linearization related to non-Gaussian and nonlinear state estimations caused by Extended Kalman Filter. A group of sigma points in UKF are used to approximate system statistical properties, in which sigma points are calculated based on nonlinear equation instead of linear one, for the proposed method has many superior advantages as compared to the traditional one. Finally the metric map is built and the superior performances of the proposed method are shown in experiments.